This goal of this repository was to minimize the number of code edits by enabling easy configuration of the Image Classifier pipeline using Hydra, Timm & Lightning. In the /config/dataset
directory, we are able to configure the data transformation without having to edit the code. In the /config/pytorch-lightning
directory, we are able to configure over 700 State-of-the-art CNN model & 10 optimizers supported by Timm. In the /config/training
directory, variables like max_epochs
can be set.
The structure of the configuration folder is shown below. The main configuration file can be found at /config/config.yaml
.
├── config
│ ├── dataset
│ ├── pytorch-lightning <- models & optimizers
│ ├── testing
│ ├── training
│ |
│ ├── configs.yaml <- main config
- Install packages
pip install -r requirements.txt
- Change the configs at
config/dataset/dataset.yaml
, etc.data_loader: train_dataset_path: PATH_TO_YOUR_DATASET val_dataset_path: PATH_TO_YOUR_DATASET
- Launch Training
python train.py
Chau Yuan Qi - @chauyuanqi - yuanqichau@gmail.com